ScholarWorks@성균관대학교

초록

Conventional AI assistants require households to choose between convenience and privacy by relying on manual account switching or shared profiles. This paper presents Famigo, a voice assistant framework that balances personal privacy with family-level information sharing. Famigo employs a parallel-repository architecture that separates private memories from shared knowledge. In addition, the framework integrates automatic user identification, lightweight on-device voice processing, and cloud-based language understanding and retrievalaugmented generation. The prototype demonstrates effective privacy preservation and natural conversational responsiveness, achieving an average end-to-end latency of 2-3 seconds. Famigo provides a foundation for developing privacy-conscious multiuser voice assistants suitable for domestic environments.

키워드